Google DeepMind announced on November 10 that its AlphaProof Code system had mathematically verified key security properties of the Signal protocol's implementation — the first time any method, with or without AI, has done so at this depth, according to the company1.
The Signal protocol secures private communications for billions of users in apps including Signal and Google Messages. Cryptographic flaws often hide in the gap between a protocol's design specification and its compiled code, a gap that prior formal verification work has not bridged for a widely deployed production protocol.
Formal verification — proving mathematically that code behaves as specified — is considered the highest standard of software security assurance but has historically required years of expert effort. Automatica reported earlier this week2 on a preprint arguing that code has become the core operational substrate for AI agents, a framing consistent with DeepMind's multi-agent approach here.
AlphaProof Code is a multi-LLM agentic framework in which human experts and AI models jointly define what must be proved, then agents powered by Gemini Carbon generate and audit the proofs, DeepMind said1. The system produced more than six million lines of Lean, a formal proof language. Human reviewers were directed to just one percent of the codebase, with Lean checking the remainder automatically. The full process took four weeks.
To test the system's ability to distinguish correct from flawed code, the team reintroduced two previously fixed bugs into an isolated copy of Signal's code; AlphaProof Code produced machine-checked proofs that both buggy versions were insecure.
DeepMind said it is open-sourcing the verification artifacts and has published a technical paper. The announcement describes no independent external audit of the methodology. The blog post is a company statement; the linked technical paper's peer-review status is not established in the available material.